Entry III · Timekeeping
Putting a pendulum in beat by count rather than by guess
In beat does not mean the tick sounds even to you across the room. It means the escapement receives the same impulse on both sides of the swing. Those two things overlap often enough to be misleading and differ often enough to waste an evening.
- Written
- 11 March 2025
- Corrected
- 6 November 2025
- Length
- 10 minutes
- Bench
- Lyon
Even to the ear is not the same as equal
The human ear is good at rhythm and bad at small asymmetry, particularly when one of the two sounds is slightly louder. A recoil escapement that is a degree out of beat will often still produce a tick and a tock that sound regular from the doorway, because the interval error is a few hundredths of a second and the difference you can hear is in the loudness rather than the timing. So a clock passes the doorway test and then stops overnight, which is what out of beat actually costs you.
The real test is whether the pendulum receives the same amount of push on the way out and the way back. If it does not, the swing is asymmetric, the escapement is relying on a larger arc than it should to unlock on one side, and the first time the arc drops a little, from a dry pivot or a slightly out-of-level case, it fails to unlock and the clock stops with the pendulum still swinging feebly.
I check it by watching rather than listening. With the dial off and the pendulum swinging small, I watch the pallets against the escape wheel teeth and see whether drop occurs at the same visible point on each side. If I cannot see it clearly, I count: I stop the pendulum, release it from exactly the position where the escapement just unlocks on the left, then from the same point on the right, and compare how far it goes each time.
Find the true centre before touching the crutch
Before bending anything, I want to know where the pendulum actually hangs at rest with the movement in its case and the case standing as it will stand. A crutch bent to correct a case that is sitting on a sloping shelf is a repair that only works on that shelf. So the case goes level, checked with a small spirit level across the seatboard rather than across the top of the case, since case tops are rarely square to the movement.
Then I let the pendulum come to rest of its own accord and mark where the bob sits against a paper scale taped behind it, in millimetres. That mark is the true hanging centre. Now I lift the crutch by hand very slowly to the left until the escapement clicks, note the reading, then slowly right until it clicks, and note that. On a well set movement those two readings are the same distance from centre, usually within a millimetre on a pendulum a third of a metre long.
On the last wall movement I had in, the left unlock was at 6 mm from centre and the right at 13 mm. That is a long way out. Somebody had fitted a new suspension spring and set the beat by ear on a wall that was not vertical, and the clock had stopped every third night since.
Bending the crutch, or not bending it
If the movement has a friction-fitted crutch or an adjustable beat block, nothing gets bent at all. The crutch is eased around its arbor a degree at a time with the pendulum off, then the readings are taken again. That is the whole job and it takes ten minutes. Where the crutch is a solid wire, it has to be bent, and bending it in the middle of its length is how you end up with a crutch that is springy and swallows part of the impulse.
I bend close to the crutch arbor, holding the wire firmly on both sides of the bend with two pairs of smooth-jawed pliers so that no twist gets introduced, and I bend about half the amount I think is needed. The reason is that a wire bent twice in the same direction is fine, while a wire bent past the mark and brought back is work-hardened at exactly the point that takes the load. I have broken one crutch that way and made another that rang like a tuning fork.
After each adjustment the pendulum goes back, the readings get retaken, and I write both numbers down. On that wall movement it took three passes to get from six and thirteen to nine and ten, which is close enough that the clock will survive a wall that is not perfect.
Why it keeps time on my board and not on your wall
This is the conversation I have most often. A movement runs on my test board for a fortnight within a few seconds a week and then loses four minutes in the first week at home. Almost none of those cases are a fault in the movement. My board is a solid oak plank bolted through into a masonry wall, at a steady eighteen or nineteen degrees, and it does not move when somebody shuts a door.
A clock case hanging from one screw into a plasterboard partition moves a little on every footstep and takes a fraction of the pendulum energy with it each time. The arc drops, and on a recoil escapement a smaller arc means the clock runs slightly fast, while on a dead beat it barely changes. A case standing on a bookshelf that is not quite level puts the pendulum out of beat again the day it goes home, no matter how carefully I set it here.
Temperature does the rest. A steel pendulum rod a third of a metre long grows enough between a January hallway at fourteen degrees and a July one at twenty-six to change the rate by several seconds a day. That is not a repair, it is physics, and the honest answer is that the rating nut will need a small turn twice a year. I write the expected direction of that on the job sheet so nobody thinks the clock has broken again.
Regulating by the nut, and how long to wait
Once the beat is right, rate is a separate and much slower job. I set the movement running against a clock I trust, write the error at the same hour each evening, and leave it alone for at least four days before touching the nut. Adjusting daily produces a clock that is always chasing yesterday, because an eight-day spring movement is naturally faster at the start of its wind than at the end.
The arithmetic is simple enough to do on paper. Raising the bob shortens the effective pendulum and speeds the clock. On a typical mantel movement with a rod a shade over a quarter of a metre, one full turn of the rating nut moves the bob by about half a millimetre and changes the rate by roughly twenty seconds a day. So a clock losing five seconds a day wants a quarter turn up, and then another fortnight of patience.
I record every turn in the job book with the date and the measured error before and after. Two things come out of that. The first is that I can see whether a movement is settling or drifting, which tells me if something is still wrong inside. The second is that after ten years of these entries I can look at a new movement of a familiar pattern and know roughly what the nut is worth before I start.